Anode furnace copper smelting redox end point intelligent judgment system and method

Abstract

The invention relates to an intelligent judgment system and method for an oxidation-reduction end point of copper smelting of an anode furnace. According to the system provided by the invention, copper sample images and environmental parameters are obtained through the bimodal image acquisition module, oxidation period and reduction period features are extracted and fused by utilizing the differentiation feature processing module, end point probability prediction is carried out by virtue of a hybrid neural network model, and finally, the intelligent control module adjusts the oxygen supply rate according to a prediction result and gives an alarm. Aiming at the problems of strong subjectivity, low precision, poor adaptability and the like due to the fact that end point judgment of an oxidation period and a reduction period depends on artificial experience or a single data source in a traditional copper fire refining process of the anode furnace, accurate judgment is realized through the steps of synchronous data acquisition, feature extraction, end point joint prediction, closed-loop control and the like; the method has the advantages of cost reduction, efficiency improvement, accurate judgment, high adaptability and the like, expensive equipment is omitted, the cost is reduced, the judgment accuracy is improved through multi-dimensional analysis, and the unique design adapts to complex working conditions.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent control of metallurgical processes, and in particular to a system and method for intelligently determining the redox endpoint of copper smelting in an anode furnace. Background Art

[0002] Anode furnace refining is a core process in the copper smelting industry. It removes impurities and sulfur from the copper anode through high-temperature reactions such as oxidation and reduction to produce high-purity copper. With increasing demands for copper purity in sectors such as new energy and electronics, precise control of the endpoints of the oxidation and reduction phases has become crucial for improving production efficiency and product quality.

[0003] In traditional processes, endpoint determination relies primarily on manual experience or single sensor data (such as temperature or current). Manual experience is significantly influenced by operator subjective factors and can easily lead to misjudgment. Automated methods based on single sensors (such as temperature threshold alarms) fail to account for changes in the physical and chemical state of the copper sample, making them difficult to adapt to fluctuations in raw material composition or environmental disturbances within the furnace, resulting in insufficient accuracy and robustness.

[0004] In judging the end point of the oxidation period, existing technologies mostly use surface color feature analysis of copper samples. However, color changes are not only related to the degree of oxidation, but are also affected by factors such as the distribution of impurities in the raw materials and the local temperature gradient in the furnace. For example, sulfide impurities may cause local abnormal color development, and uneven airflow in the furnace can easily cause differences in color distribution, resulting in a false alarm rate of up to 15%-20% for the traditional color threshold method. For judging the end point of the reduction period, existing technologies focus on grayscale texture analysis (such as grayscale co-occurrence matrix), but ignore two key issues: First, the surface texture is easily affected by environmental factors such as slag adhesion and uneven illumination, and the texture feature stability is poor under actual working conditions; second, it is not combined with changes in cross-sectional morphology (such as pore closure and grain boundary reconstruction), resulting in an inability to fully characterize the reduction reaction process.

[0005] In addition, existing technologies have systemic flaws:

[0006] 1. Single data dimension: The oxidation and reduction periods rely on single-modal data of color or texture, respectively, lacking multi-view (surface / cross-section) feature complementarity, resulting in serious loss of key information;

[0007] 2. Lack of cross-cycle synergy: The residual pore structure during the oxidation period directly affects the reduction kinetics, but the existing system does not establish an oxidation-reduction parameter correlation model, resulting in a lag in process parameter adjustment;

[0008] 3. Insufficient model adaptability: Traditional machine learning models (such as SVM and random forest) rely on fixed feature engineering and are unable to cope with fluctuations in raw material composition (such as changes in the proportion of arsenic-containing copper ore) or sudden changes in operating conditions (such as oxygen supply pressure fluctuations of ±20%).

[0009] In recent years, multi-sensor fusion solutions (such as spectrometers and infrared thermal imagers) have partially addressed the data integrity issue. However, their high hardware costs (over 500,000 yuan per unit) and complex maintenance (requiring regular calibration and dust removal) have limited their industrial adoption. Therefore, there is an urgent need for an intelligent endpoint determination system based on low-cost image sensing, integrating multi-dimensional topographic features, and possessing adaptive capabilities to achieve efficient, accurate, and cost-effective refining process control. Summary of the Invention

[0010] In order to solve or partially solve the problems existing in the relevant technologies, the present application provides an intelligent judgment system and method for the redox endpoint of copper smelting in an anode furnace, which has the advantages of cost reduction and efficiency improvement, accurate judgment, and strong adaptability. It eliminates expensive equipment to reduce costs, improves judgment accuracy through multi-dimensional analysis, and has a unique design that adapts to complex working conditions.

[0011] The present application provides a method for intelligently determining the redox endpoint of copper smelting in an anode furnace, comprising the following steps:

[0012] Step 1: Configure a high-resolution microscope camera to capture cross-sectional images of the copper sample at a sampling frequency of 10-15 frames per second during the oxidation period, and simultaneously capture surface and cross-sectional images during the reduction period. The image coverage area should be ≥80% of the copper sample surface area.

[0013] Step 2: Perform dynamic threshold segmentation in HSV color space on the oxidation period cross-section image, and calculate the porosity by combining morphological opening operation; extract texture contrast and topographic smoothness features from the reduction period surface and cross-section images respectively.

[0014] Step 3: Construct an oxidation period temporal convolutional network and a reduction period dual-path CNN model to process color feature sequences and texture-morphology features respectively, and realize multimodal feature fusion through a fully connected layer.

[0015] Step 4: Based on the nonlinear mapping relationship between the porosity in the oxidation period and the flatness in the reduction period, the prediction results are dynamically corrected; the integrated fuzzy PID controller adjusts the process parameters in real time and triggers a three-level alarm mechanism.

[0016] Step 5: When an abnormal operating condition is detected, switch to the fuzzy rule base control mode and dynamically update the model through the digital twin model and incremental learning strategy.

[0017] Optionally, in some schemes, when collecting surface and cross-sectional images during the reduction period, dual-light source multi-angle illumination is used, with the main light source having a wavelength of 550nm and the auxiliary light source having a wavelength of 850nm, to enhance the surface texture and cross-sectional morphology characteristics, respectively.

[0018] Optionally, in some embodiments, the oxidation period feature extraction method in step 2 is:

[0019] In the dynamic threshold segmentation algorithm, the H channel threshold is calculated according to the formula

[0020] H new =0.7H hist +0.3H current

[0021] Online update, where H hist is the historical mean, H current The statistical value of the current frame.

[0022] The porosity calculation uses an adaptive binarization algorithm:

[0023]

[0024] Among them I pore It is generated by combining the Otsu algorithm with the region growing method, and the threshold of the pore connected domain area is ≥50 pixels.

[0025] Optionally, in some embodiments, the reduction period feature extraction method in step 2 includes:

[0026] The surface texture contrast is calculated using the improved gray-level co-occurrence matrix (GLCM) energy-weighted formula:

[0027]

[0028] Where P(i, j) is the gray-level co-occurrence matrix probability value, and the contrast threshold is set to 120±5% (confidence interval 95%).

[0029] The cross-sectional morphology flatness index was tested by using a multi-scale Canny operator (σ = 1.0, 2.0, 3.0) to detect the edge curvature, and the curvature radius difference threshold was ≤ 5 μm.

[0030] Optionally, in some solutions, in step 4, the cross-cycle correction module dynamically corrects the prediction result based on the nonlinear mapping relationship between the porosity in the oxidation period and the flatness in the reduction period. The cross-cycle correction formula is:

[0031] P red =P pred +0.12·ln(1+Δporosity)

[0032] The Δ porosity is the difference between the porosity in the oxidation period and the porosity in the reduction period. The corrected prediction probability P red Used for adjusting the parameters of fuzzy PID controller, with a proportional band of ±8%, an integral time of 60s, and a differential time of 15s.

[0033] Optionally, in some solutions, the model self-optimization strategy adopted in step 5 includes:

[0034] When the standard deviation of the image features of three consecutive frames exceeds the 2σ range of the historical mean, it automatically switches to the fuzzy rule base control mode based on air supply pressure and oxygen potential.

[0035] Establish a digital twin model and integrate metallurgical dynamics equations:

[0036]

[0037] where k1 = 0.05·e 0.1·孔隙率 , k2 is the oxidation reaction decay constant.

[0038] When the coefficient of variation (CV) of the prediction results is greater than 15%, an incremental learning strategy is used to update the network weights, retaining key frames with feature change rates greater than 10% / min, and compressing the model size to 70% of its original size through knowledge distillation.

[0039] An intelligent determination system for the redox endpoint of copper smelting in an anode furnace comprises:

[0040] Dual-modal image acquisition module: This is the high-resolution microscope camera in step 1, which is used to acquire cross-sectional images of copper samples during the oxidation period and simultaneously acquire surface and cross-sectional images during the reduction period.

[0041] Differentiation feature processing module: applied to step 2, the HSV color space dynamic threshold segmentation algorithm is used in the oxidation period to extract the cross-section color features and calculate the porosity by combining the morphological opening operation; the H channel threshold is based on H new =0.7H hist +0.3H current Online update; During the reduction period, the surface texture contrast is calculated based on the gray-level co-occurrence matrix, and the multi-scale Canny operator is used to detect the cross-section edge curvature to generate the topography flatness index.

[0042] Hybrid neural network model: Applied to steps 3 and 4, the oxidation branch adopts a temporal convolutional network, which contains 4 layers of dilated convolution. The input is a sequence of color features of 10 consecutive frames of cross-sectional images. The loss function is: L = α·MSE+(1-α)·KLDivergence, where α = 0.8-0.02×epoch; the reduction branch consists of a two-way CNN. The first one uses ResNet18 to extract surface texture features, and the second one uses UNet to extract cross-sectional morphology features. The dual-way output is fused through a fully connected layer.

[0043] Intelligent control module: acts on steps 4 and 5, integrates fuzzy PID controller, and calculates the predicted probability value P ox >95% or P red >90% real-time adjustment of oxygen delivery rate, and equipped with a three-level alarm unit.

[0044] Optionally, in some schemes, the high-resolution microscope camera is installed at an inclination angle of 45°±5°, and a checkerboard calibration method is used to eliminate perspective distortion when collecting oxidation period cross-sectional images. The reprojection error is controlled within ±0.3 pixels. The high-resolution microscope camera is equipped with an active polarization filter and a self-cleaning air curtain system, and the self-cleaning air curtain system adopts a ring nozzle design.

[0045] Optionally, in some schemes, the intelligent control module includes an abnormal operating condition processing unit. When the standard deviation of the image features of three consecutive frames exceeds the 2σ range of the historical mean, it automatically switches to the fuzzy rule base control mode based on the supply air pressure and oxygen potential, and triggers the equipment self-test program.

[0046] The technical solution provided by this application may have the following beneficial effects:

[0047] This application addresses the problems of relying on manual experience or a single data source to judge the endpoints of the oxidation and reduction periods in the traditional anode furnace fire-based copper refining process, which has strong subjectivity, low precision, and poor adaptability. It achieves accurate judgment through steps such as synchronous data acquisition, feature extraction, joint prediction of endpoints, and closed-loop control. It has the advantages of cost reduction and efficiency improvement, accurate judgment, and strong adaptability. It eliminates expensive equipment to reduce costs, improves judgment accuracy through multi-dimensional analysis, and has a unique design that adapts to complex working conditions. Through multimodal image fusion and adaptive algorithms, the false alarm rate of endpoint judgment is reduced from 15%-20% of traditional methods to below 5%. At the same time, hardware costs are reduced by 60%, and energy consumption is reduced by 10%-15%, significantly improving the economy and reliability of the anode furnace copper smelting process.

[0048] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The above and other objects, features and advantages of the present application will become more apparent through a more detailed description of exemplary embodiments of the present application in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments of the present application.

[0050] Figure 1 This is a structural block diagram of an intelligent system for determining the redox endpoint of copper smelting in an anode furnace, as shown in an embodiment of the present application;

[0051] Figure 2 This is a schematic diagram of the installation structure of the intelligent system for determining the redox endpoint of copper smelting in an anode furnace, as shown in an embodiment of the present application;

[0052] Figure 3 This is an HSV gray histogram threshold effect diagram of the oxidation period cross-sectional image shown in the embodiment of the present application;

[0053] Figure 4This is a GLCM distribution diagram of the surface texture during the reduction period shown in the examples of the present application;

[0054] Figure 5 This is a detailed diagram of the hybrid neural network structure shown in an embodiment of the present application.

[0055] Reference numerals:

[0056] 1- Main control room connecting line, 2- Anode furnace body, 3- Rotating gear, 4- Flue gas outlet, 5- Hydraulic cylinder, 6- Feed inlet cover, 7- Oxidant and reducing agent delivery pipelines, 8- Copper outlet, 9- High-resolution microscope camera, 10- Main control room host. DETAILED DESCRIPTION

[0057] The following describes embodiments of the present application in more detail with reference to the accompanying drawings. Although the accompanying drawings illustrate embodiments of the present application, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. Rather, these embodiments are provided to make the present application more thorough and complete, and to fully convey the scope of the present application to those skilled in the art.

[0058] It should be understood that although the terms "first", "second", "third", etc. may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of this application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0059] In the description of this application, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on this application.

[0060] Unless otherwise expressly specified or limited, terms such as "mounted," "connected," "connect," and "fixed" should be interpreted broadly. For example, they may refer to fixed or detachable connections, or integration; mechanical or electrical connections; direct or indirect connections through an intermediary; and internal communication between two components or interaction between two components. Those skilled in the art will understand the specific meanings of these terms in this application based on the specific circumstances.

[0061] In response to the above problems, the embodiments of the present application provide an intelligent system and method for determining the redox endpoint of copper smelting in an anode furnace, which has the advantages of cost reduction and efficiency improvement, accurate judgment, and strong adaptability. It eliminates expensive equipment to reduce costs, improves judgment accuracy through multi-dimensional analysis, and has a unique design that adapts to complex working conditions.

[0062] The technical solutions of the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0063] See also Figure 1 The method for intelligently determining the redox endpoint of copper smelting in an anode furnace comprises the following steps:

[0064] Step 1: Configure a high-resolution microscope camera to capture cross-sectional images of the copper sample at a sampling frequency of 10-15 frames per second during the oxidation period, and simultaneously capture surface and cross-sectional images during the reduction period. The image coverage area should be ≥80% of the copper sample surface area.

[0065] Step 2: Perform dynamic threshold segmentation in HSV color space on the oxidation period cross-section image, and calculate the porosity by combining morphological opening operation; extract texture contrast and topographic smoothness features from the reduction period surface and cross-section images respectively.

[0066] Step 3: Construct an oxidation period temporal convolutional network and a reduction period dual-path CNN model to process color feature sequences and texture-morphology features respectively, and realize multimodal feature fusion through a fully connected layer.

[0067] Step 4: Based on the nonlinear mapping relationship between the porosity in the oxidation period and the flatness in the reduction period, the prediction results are dynamically corrected; the integrated fuzzy PID controller adjusts the process parameters in real time and triggers a three-level alarm mechanism.

[0068] Step 5: When an abnormal operating condition is detected, switch to the fuzzy rule base control mode and dynamically update the model through the digital twin model and incremental learning strategy.

[0069] The high-resolution microscope camera has a resolution of ≥5μm / pixel, and a digital twin model of the redox process is established to visualize the porosity evolution, texture contrast distribution, and predicted endpoint time in real time. The incremental learning strategy method is: when the coefficient of variation (CV) of the system's prediction results is >15% for five consecutive times, the model retraining process is automatically started to update the network weight parameters.

[0070] In some embodiments, when collecting surface and cross-sectional images during the reduction period, dual-light source multi-angle illumination is used, with a main light source wavelength of 550nm and an auxiliary light source wavelength of 850nm to enhance surface texture and cross-sectional morphology characteristics, respectively.

[0071] In some embodiments, the oxidation period feature extraction method in step 2 is:

[0072] In the dynamic threshold segmentation algorithm, the H channel threshold is calculated according to the formula

[0073] H new =0.7H hist +0.3H current

[0074] Online update, where H hist is the historical mean, H current The statistical value of the current frame.

[0075] The porosity calculation uses an adaptive binarization algorithm:

[0076]

[0077] Among them I pore It is generated by combining the Otsu algorithm with the region growing method, and the threshold of the pore connected domain area is ≥50 pixels.

[0078] In some embodiments, the reduction period feature extraction method in step 2 includes:

[0079] The surface texture contrast is calculated using the improved gray-level co-occurrence matrix (GLCM) energy-weighted formula:

[0080]

[0081] Where P(i, j) is the gray-level co-occurrence matrix probability value, and the contrast threshold is set to 120±5% (confidence interval 95%).

[0082] The cross-sectional morphology flatness index was tested by using a multi-scale Canny operator (σ = 1.0, 2.0, 3.0) to detect the edge curvature, and the curvature radius difference threshold was ≤ 5 μm.

[0083] In some embodiments, in step 4, the cross-cycle correction module dynamically corrects the prediction result based on the nonlinear mapping relationship between the porosity in the oxidation period and the flatness in the reduction period. The cross-cycle correction formula is:

[0084] P red =P pred +0.12·ln(1+Δporosity)

[0085] The Δ porosity is the difference between the porosity in the oxidation period and the porosity in the reduction period. The corrected prediction probability P red Used for adjusting the parameters of fuzzy PID controller, with a proportional band of ±8%, an integral time of 60s, and a differential time of 15s.

[0086] In some embodiments, the model self-optimization strategy adopted in step 5 includes:

[0087] When the standard deviation of the image features of three consecutive frames exceeds the 2σ range of the historical mean, it automatically switches to the fuzzy rule base control mode based on air supply pressure and oxygen potential.

[0088] Establish a digital twin model and integrate metallurgical dynamics equations:

[0089]

[0090] where k1 = 0.05·e 0.1·孔隙率 , k2 is the oxidation reaction decay constant.

[0091] When the coefficient of variation (CV) of the prediction results is greater than 15%, an incremental learning strategy is used to update the network weights, retaining key frames with feature change rates greater than 10% / min, and compressing the model size to 70% of its original size through knowledge distillation.

[0092] An intelligent determination system for the redox endpoint of copper smelting in an anode furnace comprises:

[0093] Dual-modal image acquisition module: This is the high-resolution microscope camera in step 1, which is used to acquire cross-sectional images of copper samples during the oxidation period and simultaneously acquire surface and cross-sectional images during the reduction period.

[0094] Differentiation feature processing module: applied to step 2, the HSV color space dynamic threshold segmentation algorithm is used in the oxidation period to extract the cross-section color features and calculate the porosity by combining the morphological opening operation; the H channel threshold is based on H new =0.7H hist +0.3H current Online update; During the reduction period, the surface texture contrast is calculated based on the gray-level co-occurrence matrix, and the multi-scale Canny operator is used to detect the cross-section edge curvature to generate the topography flatness index.

[0095] The threshold for determining the pore area was grayscale value <50. During the reduction period, the surface texture contrast was calculated based on the grayscale co-occurrence matrix (window size 32 × 32 pixels, step size 8 pixels). At the same time, the multi-scale Canny operator (σ = 1.0, 2.0, 3.0) was used to detect the curvature of the cross-section edge and generate the topographic flatness index.

[0096] Hybrid neural network model: Applied to steps 3 and 4, the oxidation branch adopts a temporal convolutional network, which contains 4 layers of dilated convolution. The input is a sequence of color features of 10 consecutive frames of cross-sectional images. The loss function is: L = α·MSE+(1-α)·KLDivergence, where α = 0.8-0.02×epoch; the reduction branch consists of a two-way CNN. The first one uses ResNet18 to extract surface texture features, and the second one uses UNet to extract cross-sectional morphology features. The dual-way output is fused through a fully connected layer.

[0097] Intelligent control module: acts on steps 4 and 5, integrates fuzzy PID controller, and calculates the predicted probability value P ox >95% or P red >90% real-time adjustment of oxygen delivery rate, and equipped with a three-level alarm unit.

[0098] Integrated fuzzy PID controller (proportional band ± 8%, integral time 60s, differential time 15s), according to the predicted probability value P ox >95% or P red >90% real-time adjustment of oxygen delivery rate, and equipped with three-level alarm unit (early warning, emergency alarm, shutdown protection).

[0099] See also Figure 2 A flue gas outlet 4 is provided at the top left side of the anode furnace body 2, a rotating gear 3 is provided on the outer wall of the anode furnace body 2, a feed port cover 6 is installed on the top of the anode furnace body 2 through a hydraulic cylinder 5, and oxidant and reducing agent delivery pipes 7 are installed. The oxidant and reducing agent delivery pipes 7 are connected to the inner cavity of the anode furnace body 2, and a copper outlet 8 is provided on the right side of the anode furnace body 2. A high-resolution microscope camera 9 is provided on the side of the lower part of the copper outlet 8. The high-resolution microscope camera 9 is connected to the main control room host 10 through the main control room connection line 1. The differential feature processing module, the hybrid neural network model, and the intelligent control module are electrically connected to the main control room host 10. The dual-modal image acquisition module includes a high-resolution microscope camera 9 and an image processing submodule. The high-resolution microscope camera 9 is used to collect high-definition copper sample texture images. The image processing submodule is used to process the high-definition copper sample texture images and transmit the high-definition images through the main control room connection line 1.

[0100] In some embodiments, the high-resolution microscope camera is installed at an inclination angle of 45°±5°, and a checkerboard calibration method is used to eliminate perspective distortion when collecting oxidation period cross-sectional images. The reprojection error is controlled within ±0.3 pixels. The high-resolution microscope camera is equipped with an active polarization filter and a self-cleaning air curtain system, and the self-cleaning air curtain system adopts a ring nozzle design.

[0101] The active polarizing filter has a wavelength range of 450-650nm, and perspective distortion correction uses a homography matrix transformation based on checkerboard calibration. The self-cleaning air curtain system has a nitrogen purity of ≥99.5% and a flow rate of 0.5m / s to eliminate reflective interference on the molten pool surface and prevent lens contamination. The nitrogen flow rate is controlled by a PID closed loop to ensure that the lens contamination rate is less than 0.1% / hour.

[0102] In some embodiments, the intelligent control module includes an abnormal operating condition processing unit. When the standard deviation of the image features of three consecutive frames exceeds the 2σ range of the historical mean, it automatically switches to the fuzzy rule base control mode based on the air supply pressure and oxygen potential, and triggers the equipment self-test program.

[0103] Working process of this application:

[0104] See also Figure 3-5 A high-resolution microscope camera (resolution ≥ 5μm / pixel) is installed at the anode furnace observation window, equipped with an active polarization filter (wavelength range 450-650nm) to suppress reflection interference from the molten pool surface. The camera is installed at an inclination angle of 45°±5°, and a checkerboard calibration method is used to eliminate perspective distortion, ensuring a reprojection error of ≤±0.3 pixels. The self-cleaning air curtain system uses an annular nitrogen nozzle (purity ≥ 99.5%). The flow rate is stabilized at 0.5m / s through PID closed-loop control, keeping the lens contamination rate less than 0.1% / hour.

[0105] During the oxidation phase, cross-sectional images of the copper samples were acquired at a sampling rate of 10-15 frames per second, with a single frame covering ≥80% of the sample surface area. During the reduction phase, surface and cross-sectional images were acquired simultaneously, using dual-light source multi-angle illumination (primary light source 550nm, auxiliary light source 850nm) to enhance texture and morphology.

[0106] The cross-sectional image was converted to the HSV color space, and the H channel threshold was dynamically updated according to the formula. The adaptive binarization algorithm (Otsu+region growing method) was used to generate the pore area mask, and the pore connected domain area threshold was set to ≥50 pixels.

[0107] Contrast was calculated based on a modified gray-level co-occurrence matrix (GLCM) with a window size of 32 × 32 pixels, a step size of 8 pixels, and a contrast threshold of 120 ± 5% (95% confidence interval). Edge curvature was detected using a multiscale Canny operator (σ = 1.0, 2.0, 3.0), with a curvature radius difference threshold of ≤ 5 μm.

[0108] The oxidation branch (TCN network) inputs a sequence of color features from 10 consecutive cross-sectional images. The network contains four layers of dilated convolutions (with dilation coefficients of [1, 2, 4, 8]). The loss function uses a time decay factor to balance prediction error and distribution consistency.

[0109] In the restoration branch (dual-path CNN), the ResNet18 branch takes a surface image (256×256×3) as input, extracts texture features using a residual block, and outputs a 512-dimensional vector. The UNet branch takes a cross-sectional image (512×512×3) as input, uses an encoder-decoder structure to preserve spatial details, and outputs a 512-dimensional vector.

[0110] The dual outputs are then concatenated into a 1024-dimensional vector and mapped to the target category (not yet reached / end point) through a fully connected layer.

[0111] Finally, it should be noted that, in this document, relationships such as first and second, etc., are used solely to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms include, comprise, or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0112] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0113] The embodiments of the present application have been described above. The above description is illustrative and not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or improvements to the technology in the market, or to enable other persons skilled in the art to understand the embodiments disclosed herein.

Claims

1. An intelligent method for determining the redox endpoint of copper smelting in an anode furnace, characterized by: The method for intelligently determining the redox endpoint of copper smelting in an anode furnace comprises the following steps: Step 1: Configure a high-resolution microscope camera to capture cross-sectional images of the copper sample at a sampling frequency of 10-15 frames per second during the oxidation period, and simultaneously capture surface and cross-sectional images during the reduction period. The image coverage area should be ≥80% of the copper sample surface area. Step 2: Perform dynamic threshold segmentation of the oxidation period cross-section image using HSV color space, and calculate the porosity by combining morphological opening operation; extract texture contrast and topographic smoothness features from the reduction period surface and cross-section images respectively; Step 3: Construct an oxidation-period temporal convolutional network and a reduction-period dual-path CNN model to process color feature sequences and texture-topography features respectively, and implement multimodal feature fusion through a fully connected layer; Step 4: Based on the nonlinear mapping relationship between porosity during oxidation and flatness during reduction, the prediction results are dynamically corrected. The integrated fuzzy PID controller adjusts the process parameters in real time and triggers a three-level alarm mechanism. Step 5: When an abnormal operating condition is detected, switch to the fuzzy rule base control mode and dynamically update the model through the digital twin model and incremental learning strategy.

2. The intelligent method for determining the redox endpoint of copper smelting in an anode furnace according to claim 1, characterized in that: When collecting surface and cross-sectional images during the reduction period, dual light sources with multi-angle illumination are used, with a main light source wavelength of 550nm and an auxiliary light source wavelength of 850nm, to enhance the surface texture and cross-sectional morphology characteristics respectively.

3. The intelligent method for determining the redox endpoint of copper smelting in an anode furnace according to claim 2, characterized in that: The oxidation period feature extraction method in step 2 is: In the dynamic threshold segmentation algorithm, the H channel threshold is calculated according to the formula <h2 style=";text-align:left;direction:ltr">H<h2 style=";text-align:left;direction:ltr"> new <h2 style=";text-align:left;direction:ltr"> <0.7H<h2 style=";text-align:left;direction:ltr"> hist <h2 style=";text-align:left;direction:ltr"> +0.3H<h2 style=";text-align:left;direction:ltr"> current Online update, where H hist is the historical mean, H current is the current frame statistics; The porosity calculation uses an adaptive binarization algorithm: Among them I pore It is generated by combining the Otsu algorithm with the region growing method, and the threshold of the pore connected domain area is ≥50 pixels.

4. The intelligent method for determining the redox endpoint of copper smelting in an anode furnace according to claim 3, characterized in that: The reduction period feature extraction method in step 2 includes: The surface texture contrast is calculated using the improved gray-level co-occurrence matrix (GLCM) energy-weighted formula: Where P(i,j) is the gray-level co-occurrence matrix probability value, and the contrast threshold is set to 120±5% (95% confidence interval); The cross-sectional morphology flatness index was tested by using a multi-scale Canny operator (σ = 1.0, 2.0, 3.0) to detect the edge curvature, and the curvature radius difference threshold was ≤ 5 μm.

5. The intelligent method for determining the redox endpoint of copper smelting in an anode furnace according to claim 4, characterized in that: In the fourth step, the cross-cycle correction module dynamically corrects the prediction results based on the nonlinear mapping relationship between the porosity in the oxidation period and the flatness in the reduction period. The cross-cycle correction formula is: P red =P pred +0.12·ln(1+Δporosity) The Δ porosity is the difference between the porosity in the oxidation period and the porosity in the reduction period. The corrected prediction probability P red Used for adjusting the parameters of fuzzy PID controller, with a proportional band of ±8%, an integral time of 60s, and a differential time of 15s.

6. The intelligent method for determining the redox endpoint of copper smelting in an anode furnace according to claim 5, characterized in that: The model self-optimization strategy adopted in step 5 includes: When the standard deviation of the image features of three consecutive frames exceeds the 2σ range of the historical mean, it automatically switches to the fuzzy rule base control mode based on air supply pressure and oxygen potential; Establish a digital twin model and integrate metallurgical dynamics equations: where k1 = 0.05·e 0.1·孔隙率 , k2 is the oxidation reaction decay constant; When the coefficient of variation (CV) of the prediction results is greater than 15%, an incremental learning strategy is used to update the network weights, retaining key frames with feature change rates greater than 10% / min, and compressing the model size to 70% of its original size through knowledge distillation.

7. A judgment system based on the intelligent judgment method for the redox endpoint of copper smelting in an anode furnace according to any one of claims 1 to 6 comprises: Dual-modal image acquisition module: This is the high-resolution microscope camera in step 1, which is used to acquire cross-sectional images of the copper sample during the oxidation period and simultaneously acquire surface and cross-sectional images during the reduction period; Differentiation feature processing module: applied to step 2, the HSV color space dynamic threshold segmentation algorithm is used in the oxidation period to extract the cross-section color features and calculate the porosity by combining the morphological opening operation; the H channel threshold is based on H new =0.7H hist +0.3H current Online update; During the reduction period, the surface texture contrast is calculated based on the gray-level co-occurrence matrix, and the multi-scale Canny operator is used to detect the edge curvature of the cross section to generate the topography flatness index; Hybrid neural network model: Applied to steps 3 and 4, the oxidation branch uses a temporal convolutional network with four layers of dilated convolutions. The input is a sequence of color features of 10 consecutive cross-sectional images. The loss function is: L = α·MSE + (1-α)·KLDivergence, where α = 0.8-0.02×epoch. The reduction branch consists of a two-way CNN. The first channel uses ResNet18 to extract surface texture features, and the second channel uses UNet to extract cross-sectional morphology features. The dual-channel outputs are fused through a fully connected layer. Intelligent control module: acts on steps 4 and 5, integrates fuzzy PID controller, and calculates the predicted probability value P ox >95% or P red >90% real-time adjustment of oxygen delivery rate, and equipped with a three-level alarm unit.

8. The intelligent system for determining the redox endpoint of copper smelting in an anode furnace according to claim 7, characterized in that: The high-resolution microscope camera is installed at an inclination angle of 45°±5°. When collecting oxidation period cross-sectional images, the checkerboard calibration method is used to eliminate perspective distortion, and the reprojection error is controlled within ±0.3 pixels. The high-resolution microscope camera is equipped with an active polarization filter and a self-cleaning air curtain system. The self-cleaning air curtain system adopts a ring nozzle design.

9. The intelligent system for determining the redox endpoint of copper smelting in an anode furnace according to claim 8, characterized in that: The intelligent control module includes an abnormal operating condition processing unit. When the standard deviation of the image features of three consecutive frames exceeds the 2σ range of the historical mean, it automatically switches to the fuzzy rule base control mode based on the air supply pressure and oxygen potential, and triggers the equipment self-test program.

Images